AI decisions Operating model
Fix the data first or start with AI now: which comes first?
Do not wait for a perfect data platform before starting with AI, and do not start with AI while ignoring the data. Pick a few focused use cases, find out exactly which data each one needs and fix that data to the standard the case requires. A broad clean-up first makes sense only when the same data gaps block every valuable case, or when regulation requires data governance you do not yet have.
The options
Fix data first
A programme to consolidate, clean and govern the data platform before building AI use cases on top of it.
Start with focused AI
A small number of AI use cases started now, with data work limited to what each case needs to run well.
Side by side
| Criterion | Fix data first | Start with focused AI |
|---|---|---|
| Time to first value | Late: value arrives once the platform is ready and use cases are built on it. | Earlier: a focused case can reach users while the wider data work continues. |
| Scope of data work | Broad: domains, sources and quality rules defined for the whole organisation. | Narrow: only the sources, fields and quality the chosen cases depend on. |
| Prioritisation | Driven by data architecture; it is hard to know which fixes matter most until a use case needs them. | Driven by use cases, which show which data gaps actually affect results. |
| Risk of rework | Data modelled without real consumers may not fit the needs AI cases reveal later. | Point fixes can multiply into separate pipelines if nobody consolidates them. |
| Fit to unstructured data | Platform programmes often focus on structured data and leave documents for later. | Many generative AI cases start from documents and knowledge, which need curation, permissions and metadata rather than a warehouse. |
| Governance and compliance | Builds lineage, ownership and quality controls once, for everyone. | Governance must be applied per case from day one, or it is skipped. |
| Sponsorship | Hard to sustain a long programme with no visible result for the business. | Early results help fund the data work that follows. |
Choose Fix data first when…
- Every valuable use case you have assessed is blocked by the same data gaps.
- Core data has no clear owner, definitions conflict between areas, and nobody trusts the numbers.
- Planned AI systems may be high-risk under the EU AI Act, which sets data governance requirements for training, validation and testing data.
- A platform modernisation is already funded and AI cases can be designed as its first consumers.
Choose Start with focused AI when…
- You can name a few cases with clear value whose data is good enough or can be fixed in a bounded way.
- The first cases work mostly on documents and knowledge rather than on transactional data.
- The business needs visible results to keep sponsoring AI and data investment.
- You do not yet know which data gaps matter, and real use cases are the fastest way to find out.
When to combine them
The practical answer is usually both, in order of need. Start a few focused cases and make each one record the data it uses, the gaps it found and the fixes it made. Feed those findings into the data roadmap, so the platform work goes first to the sources several cases depend on and point fixes are consolidated instead of duplicated. Over time the use cases set the priorities of the data programme, and the programme lowers the cost of the next cases.
Common mistakes
- Waiting for a perfect, organisation-wide data platform that never quite arrives.
- Starting AI cases on data nobody owns or has checked, then blaming the model for poor results.
- Letting each case build its own pipeline and copy of the data, creating the next silo.
- Treating documents as out of scope for data work; for generative AI they are often the main source.
- Measuring data work by volume migrated instead of by the cases it unblocks.
How Thinkia approaches it
We start from the use cases, not from the data catalogue. In discovery we rank candidate cases by value and risk and, for each one, check which data it needs, who owns it and whether its quality is good enough for the decision the case supports. That tells us whether a case can start now, needs a bounded data fix first, or depends on platform work that should be planned on its own.
When the first cases rely on documents and knowledge, Enterprise Knowledge AI provides a governed knowledge layer: ingestion that keeps source permissions, access control enforced at query time and answers linked to the document and passage that support them. The work that decides quality is unglamorous: choosing sources, curating content and keeping metadata current. Synapse gives that work a governed base with corporate SSO, model routing, cost control and ROI dashboards.
We feed what each case learns about the data back into the data roadmap, so the platform work goes where several cases need it. If a planned system may be high-risk under the EU AI Act, we bring data governance forward for that case; our AI governance guide is operational guidance, not legal advice.
Thinkia products involved
- Enterprise Knowledge AIGoverned knowledge layer: source-grounded answers with citations and audit trail.
- SynapseGoverned agentic platform: agents, models, costs and data in one place.
Related AI solutions
Frequently asked questions
Do we need a data lake or warehouse before we can use AI?
Not necessarily. Many useful cases run on a small set of sources, or on documents, with the right permissions and quality checks. A consolidated platform lowers the cost of later cases, but it is not a precondition for the first ones.
How good does the data need to be?
Good enough for the decision the case supports. A draft that a person reviews tolerates more noise than an automated decision about a customer. Define the quality threshold per case and measure it, instead of aiming for perfect data everywhere.
Won't starting with use cases create more data silos?
It can, if each case builds its own pipeline in isolation. Avoid it by sharing a common platform for access and logging, recording the data each case uses and consolidating fixes into the data roadmap as they repeat.
Does the EU AI Act require clean data before deploying AI?
For high-risk AI systems, the Regulation sets data governance requirements for training, validation and testing data, including relevance, representativeness and examination for possible biases. Other systems have lighter obligations. Check the specific case against the Regulation and qualified legal advice.
Who should own the decision?
Business and data leaders together. The business brings the cases and their value; the data team brings what it costs to make the data fit for each one. A shared, prioritised list of cases and data gaps keeps both sides honest.
Keep exploring
Related decisions
- AI pilot vs production: what actually changes when you scale?
- RAG vs enterprise search: do your people need answers or documents?
- Legacy rewrite vs AI-assisted modernisation: should you start again or decode what you have?
- Synthetic data vs real data: what should you train, tune and test your AI on?
Key terms
Thinkia articles
- Data Strategy: The Foundation of AI-Driven Advantage
- Smart Data Curation Is the New Scale for Enterprise AI
- Data Fabric: Weaving a Unified View of Your Data Landscape for Strategic Advantage
- Generative AI: From Hype to Real-World Value Creation
Whitepapers
- The company with memory. Storing is not remembering.Companies do not have operational memory; they have amnesia governed by folders. A repository is not memory: real recall answers in the workflow.
- AI Dashboards. Ask, don't filter.The new business intelligence is a conversation with data, not a visualization of it—the inverted flow, three AI-native layers, and a traceable answer.